ARTHRITIS & RHEUMATISM
Vol. 63, No. 1, January 2011, pp 276–285
The Effect of Genotype on Methotrexate Polyglutamate
Variability in Juvenile Idiopathic Arthritis and
Association With Drug Response
Mara L. Becker,1 Roger Gaedigk,1 Leon van Haandel,2 Bradley Thomas,1 Andrew Lasky,1
Mark Hoeltzel,1 Hongying Dai,1 John Stobaugh,2 and J. Steven Leeder1
Objective. The response to and toxicity of metho-
trexate (MTX) are unpredictable in patients with juve-
nile idiopathic arthritis (JIA). Intracellular polygluta-
mation of MTX, assessed by measuring concentrations
of MTX polyglutamates (MTXGlu), has been demon-
strated to be a promising predictor of drug response.
Therefore, this study was aimed at investigating the
genetic predictors of MTXGlu variability and associa-
tions between MTXGlu and drug response in JIA.
Methods. The study was designed as a single-
center cross-sectional analysis of patients with JIA who
were receiving stable doses of MTX at a tertiary care
children’s hospital. After informed consent was ob-
tained from the 104 patients with JIA, blood was with-
drawn during routine MTX-screening laboratory test-
ing. Clinical data were collected by chart review.
Genotyping for 34 single-nucleotide polymorphisms
(SNPs) in 18 genes within the MTX metabolic pathway
was performed. An ion-pair chromatographic procedure
with mass spectrometric detection was used to measure
MTXGlu1–7.
Results. Analysis and genotyping of MTXGlu was
completed in the 104 patients. K-means clustering re-
Dr. Becker’s work was supported by the Katherine B. Rich-
ardson Grant, the Children’s Mercy Hospital Young Investigator
Award, the Kansas City Area Life Sciences Institute Grant, the
PhRMA Foundation Award, and the Paul Henson Immunology Re-
search Award.
1
Mara L. Becker, MD, MSCE, Roger Gaedigk, PhD, Bradley
Thomas, BA, Andrew Lasky, MD, Mark Hoeltzel, MD, Hongying Dai,
PhD, J. Steven Leeder, PharmD, PhD: Children’s Mercy Hospitals and
Clinics, Kansas City, Missouri; 2Leon van Haandel, MS, John Sto-
baugh, PhD: University of Kansas, Lawrence.
Address correspondence to Mara L. Becker, MD, MSCE,
Children’s Mercy Hospitals and Clinics, Pediatrics, Section of Rheu-
matology, 2401 Gillham Road, Kansas City, MO 64108. E-mail:
mlbecker@cmh.edu.
Submitted for publication June 4, 2010; accepted in revised
form September 30, 2010.
276
DOI 10.1002/art.30080
© 2011, American College of Rheumatology
sulted in 3 distinct patterns of MTX polyglutamation.
Cluster 1 had low red blood cell (RBC) MTXGlu
concentrations, cluster 2 had moderately high RBC
MTXGlu1 ⴙ 2 concentrations, and cluster 3 had high
concentrations of MTXGlu, specifically MTXGlu3–5.
SNPs in the purine and pyrimidine synthesis pathways,
as well as the adenosine pathway, were significantly
associated with cluster subtype. The cluster with high
concentrations of MTXGlu3–5 was associated with ele-
vated liver enzyme levels on liver function tests (LFTs),
and there were higher concentrations of MTXGlu3–5 in
children who reported gastrointestinal side effects and
had abnormal findings on LFTs. No association was
noted between MTXGlu and active arthritis.
Conclusion. MTXGlu remains a potentially useful
tool for determining outcomes in patients with JIA being
treated with MTX. The genetic predictors of MTXGlu
variability may also contribute to a better understand-
ing of the intracellular biotransformation of MTX in
these patients.
The response to methotrexate (MTX) in patients
with juvenile idiopathic arthritis (JIA) is variable, with
no identified predictors of response or toxicity (1–3).
The mechanism of action of the drug is complex and
incompletely understood. However, over the past several
decades, we have gained increasing knowledge regarding
specific sites of MTX action within the cellular folate
cycle. Due to its inhibitory effect on several target
enzymes, we know that MTX disrupts the folate cycle,
which results in decreased production of DNA precur-
sors, disruption of methyl-dependent reactions, and ac-
cumulation of adenosine, the latter of which has antiin-
flammatory properties (4–6). These resultant effects of
MTX have made it a useful therapy for conditions
dependent on rapidly dividing cells for development,
GENOTYPE EFFECTS ON MTXGlu VARIABILITY AND DRUG RESPONSE IN JIA
such as cancer and psoriasis, as well as for immuno-
inflammatory diseases such as JIA.
The polyglutamation of MTX, assessed by the
measurement of MTX polyglutamates (MTXGlu), has
been identified as a potentially useful biomarker of
treatment response in adults with rheumatoid arthritis
(RA) (7–9). The conversion of MTX intracellularly by
folylpolyglutamate synthase to its polyglutamate forms
appears to be an important process for intracellular
retention of the drug as well as for increased affinity for
its target enzymes (4,6,10,11). We have previously re-
ported the extensive variability in intracellular MTXGlu
concentrations in JIA, and have observed the effect that
route of drug administration can have on MTXGlu
concentrations (12). Due to the complexity of the folate
cycle as well as the extensive variability in response to
the drug in clinical practice, we explored potential
genetic factors that may affect distinct patterns of MTX-
Glu concentrations intracellularly, and evaluated associ-
ations between the MTXGlu concentrations and drug
response in patients with JIA.
PATIENTS AND METHODS
Patient population. This study was a cross-sectional
analysis of patients with JIA who were being evaluated at a
single center. All patients diagnosed as having JIA via Inter-
national Classification of Diseases, Ninth Revision coding
were identified, and the diagnosis was further verified through
the Edmonton 2001 International League of Associations for
Rheumatology criteria for JIA (13). Patients between the ages
of 8 months and 21 years who were receiving a stable dose and
route of MTX for a minimum of 3 months were identified for
inclusion in the study. An additional blood sample was ob-
tained at the time of a regularly scheduled blood withdrawal
for laboratory testing. Parents’ written informed consent, or
the patient’s written informed assent for those patients ages 7
years or older, was obtained from all participants, in accor-
dance with approval from the Children’s Mercy Hospital
Pediatric Institutional Review Board.
Analysis of intracellular MTX. Red blood cells (RBCs)
were washed and separated as described previously (12). To
increase specificity and minimize interference with coeluting
substances, an ion-pair reverse-phase chromatography method
was developed to measure the individual subtypes of MTXGlu
(MTXGlu1–7) in human RBCs (14). Since MTXGlu1–5 ac-
counted for 99.6% of MTXGlutotal, subsequent analyses as-
sessed MTXGlu1–5.
Genotyping method. DNA was extracted from an
aliquot of whole blood using the Illustra blood genomicPrep
Mini Spin kit (GE Healthcare). Genotyping was performed for
34 single-nucleotide polymorphisms (SNPs) in 18 genes along
the folate/MTX metabolic pathway (Table 1), using prede-
signed TaqMan assays (Applied Biosystems), sequencing reac-
tions, restriction fragment length polymorphism assays, and
277
polymerase chain reaction amplification-size discrimination
assays.
Collection of clinical data. Clinical response data
included the presence or absence of active arthritis at the time
of the visit, the number of joints with active arthritis present at
the time of the visit (defined as the active joint count), and the
erythrocyte sedimentation rate (ESR). Active arthritis was
defined as swelling of the joints not due to bony enlargement
or, if no swelling was present, limitation of motion accompa-
nied by either pain on motion or tenderness that was not due
to trauma or explained by prior joint damage (15). Addition-
ally, when charts were available for review, we retrospectively
recorded the active joint count both at 6 months and at 12
months after treatment initiation, to estimate the initial re-
sponse to MTX over time. Gastrointestinal (GI) side effects
were recorded as either 1) the presence of current GI side
effects, defined as nausea, vomiting, or abdominal pain, in
temporal relation to the weekly MTX doses reported by the
patient or parent at the time that the blood sample was
obtained, or 2) any increased folic acid supplementation (⬎1
mg/day) attributable to GI side effects in the past. In addition,
elevated liver enzyme values, defined as levels above the
normal range on liver function tests (LFTs), were recorded
during regular screening laboratory examinations, along with
the complete blood cell count (including the mean corpuscular
volume).
Additional variables that were recorded as potential
confounders included age, sex, folate use, MTX dose (in
mg/kg), months receiving treatment with MTX, route of
administration of MTX, use of nonsteroidal antiinflammatory
drugs (NSAIDs), use of anti–tumor necrosis factor ␣ (anti-
TNF␣) therapy (including etanercept, infliximab, and adali-
mumab), and use of oral corticosteroids.
Statistical analysis. Descriptive statistics were utilized
to characterize the variability in concentrations of each indi-
vidual MTXGlu subtype (MTXGlu1–5), as described previ-
ously (12). K-means clustering was performed as a class
discovery method for subgroup analyses. Bivariable associa-
tions were evaluated by chi-square test, Fisher’s exact test,
2-sample t-test, or Wilcoxon’s rank sum test, as appropriate.
Multivariate logistic regression was utilized to select significant
risk factors and their interactions, if any, from all SNPs and
covariates. Standard statistical analyses were performed using
JMP version 8 and SAS version 9.2 (both from SAS).
Multifactor dimensionality reduction (MDR) was per-
formed using the open-source software program mdr2.0, and
the model’s goodness-of-fit and significance were assessed
using mdrpt1.0 software (16). The risk factors included 34
SNPs. Cluster outcomes and combinations were assessed sep-
arately. A subset of interactive genes was first selected using
the ReliefF filtering process. High-risk, multilocus genotype
combinations, in which the ratio of cases to controls exceeded
a predetermined threshold, were selected and grouped to-
gether, which consolidated the high-dimensional risk space
into a new, 1-dimensional risk variable.
The optimal one-way, two-way, and three-way MDR
models were determined using machine-learning techniques. A
one-locus model characterized the main effect of each SNP,
while multilocus models investigated the interactions among
relevant polymorphisms. A 10-fold cross-validation procedure
randomly divided the whole data set into a training set (90% of
278 BECKER ET AL

Table 1. Genes and 34 associated single-nucleotide polymorphisms investigated within the methotrexate/folate metabolic pathway
Allele

Gene, rs no. Assay type* WT/var Frequency

ABCG2
rs7699188 Sequencing C/T 0.87/0.13
rs35252139 Sequencing G/A 0.87/0.13
rs35229708 Sequencing A/G 0.87/0.13
rs55930652 Sequencing C/T 0.73/027
ADORA2
rs2298383 TaqMan C/T 0.39/0.61
rs3761422 TaqMan C/T 0.67/0.33
rs2267076 TaqMan C/T 0.67/0.33
rs2236624 TaqMan C/T 0.77/0.23
ATIC
rs2372536 Sequencing C/G 0.70/0.30
rs4673990 TaqMan A/G 0.51/0.49
rs12995526 TaqMan C/T 0.54/0.46
BHMT
rs3733890 TaqMan G/A 0.67/0.33
DHFR
rs7387 RFLP A/T 0.70/0.30
19-nt del intron 1 PCR WT/var 0.64/0.36
GART
rs8971 RFLP A/G 0.81/0.19
TYMS
rs11280056 Sequencing WT/var (del) 0.68/0.32
rs34743033 PCR WT (*2)/var (*3)/var (*4) 0.485/0.510/0.005
GGH
rs1800909 TaqMan T/C 0.73/0.27
rs3758149 RFLP C/T 0.73/0.27
rs11545078 TaqMan C/T 0.92/0.08
ITPA
rs2295553 TaqMan T/C 0.48/0.52
MTHFD1
rs2236225 RFLP G/A 0.55/0.45
MTHFD2
rs56168672† RFLP WT (G)/var (del)/var (T) 0.57/0.31/0.12
16-nt del promoter† RFLP WT/var (del) 0.69/0.31
rs12196 RFLP A/G 0.66/0.34
MTHFR
rs1801133 TaqMan C/T 0.70/0.30
rs1801131 TaqMan A/C 0.67/0.33
rs2274976 RFLP G/A 0.94/0.06
MTR
rs1805087 RFLP A/G 0.81/0.19
MTRR
rs1801394 RFLP A/G 0.43/0.57
SHMT1
rs1979277 RFLP C/T 0.63/0.37
SLC01B1
rs2295553 TaqMan T/C 0.88/0.12
SLC19A1
rs1051266 RFLP A/G 0.45/0.55
SLC25A32
rs17803441 RFLP G/A 0.93/0.07

* RFLP ⫽ restriction fragment length polymorphism; PCR ⫽ polymerase chain reaction.

† Sequences for the wild-type (WT), variant (var), and 16-nucleotide deletion (del) were as follows: GTGTTGTCCGGCCTTCG-
GTGACACTAGC, GTGTTGTCCGTCCTTCGGTGACACTAGC (variant indicated in boldface), and GTGTTGdelACTAGC.

data) and a validation set (10% of data). The testing accuracy, data set. In addition, the cross-validation count was deter-
which is expressed as the percentage of true positive and true mined, which summarizes the number of times the optimal
negative actions, was obtained when the classification rule model developed from the training data set was confirmed by
developed from the training data set was applied to the testing the validation data set. Finally, the validity and significance of
GENOTYPE EFFECTS ON MTXGlu VARIABILITY AND DRUG RESPONSE IN JIA
Table 2. Clinical response to methotrexate within the first 12 months of treatment*
Joint count, median (range)†
SC route of administration, %
Corticosteroid use, %
NSAID use, %
Anti-TNF␣ use, %
Joint count improvement, mean ⫾ SD %
* SC ⫽ subcutaneous; NA ⫽ not applicable; NSAID ⫽ nonsteroidal antiinflammatory drug; anti-TNF␣ ⫽
anti–tumor necrosis factor ␣.
† Number of joints with active arthritis.
the selected models were assessed by 1,000-count permutation
testing.
RESULTS
Demographic characteristics of the subjects. One
hundred four patients with the underlying diagnosis of
JIA who were receiving stable doses of MTX for at least
3 months were recruited between August 2007 and
August 2009. The mean ⫾ SD age in our cohort was
117.1 ⫾ 56.5 months, and 68% were female. The major-
ity of the patients (n ⫽ 97) were Caucasian (93.2%,
including 1 of Hispanic ethnicity) and 6.7% of the
patients were African American. The disease duration in
this cohort was a median of 33.7 months (interquartile
range [IQR] 13.4–63.6 months). The median time from
diagnosis to the institution of MTX was 2.5 months
(IQR 0.89–26.9 months). The mean ⫾ SD weekly dose
of MTX was 0.51 ⫾ 0.24 mg/kg, with weekly doses
ranging from 5 mg to 25 mg (mean ⫾ SD dosage 15.9 ⫾
5.6 mg per week). The median duration of MTX therapy
was 18 months (IQR 9–46 months, range 3–156 months)
and only 44.2% of the patients reported taking supple-
mental folic acid at the time of the blood withdrawal.
Of note, 56.7% of the patients in the cohort had
active arthritis at the time of the examination (active
joint count range 1–37), and the median ESR was 12
mm/hour (range 1–105 mm/hour). Twenty-five percent
of patients reported experiencing GI side effects in
temporal relation to the MTX dose, and 14.4% had an
elevated value on LFTs (levels above the normal limits)
at the time of their concurrent screening-laboratory
blood withdrawal. All elevations in liver enzyme levels in
this cohort were ⬍2 times the upper limit of normal.
Early response to MTX. Since the study design
was cross-sectional and there were varying lengths of
MTX treatment and doses of MTX prescribed, we were
unable to utilize standard outcome measurements, such
279
Initial 6 months 12 months
(n ⫽ 93) (n ⫽ 93) (n ⫽ 90)
6 (0–63) 2 (0–48) 0 (0–42)
NA 57 68.5
17.5 18.5 9.1
81.4 87 62.2
1 18.3 23.9
NA 51.6 ⫾ 59.7 72.4 ⫾ 48.0
as the American College of Rheumatology Pediatric 30
(ACR Pedi 30), 50, and 70 criteria for improvement
(17). However, we were interested in how well our
cross-sectional outcome of active or inactive disease may
represent early response to MTX. Among the 104
patients, we collected limited clinical data, including an
initial, 6-month, and 12-month active joint count in 93
(89.4%), 93 (89.4%), and 90 (86.5%) of the patients.
The remaining clinical information was not accessible
due to the fact that the initial care of subjects was
provided at different institutions, we were unable to
retrieve the oldest records, or the 12-month data were
not yet available. The clinical characteristics of our
cohort in the first 12 months of treatment with MTX,
including the mean percentage improvement in the
active joint count at 6 months and 12 months, are
included in Table 2. There was no association between
coadministration of medication (including NSAIDs, glu-
cocorticoids, or anti-TNF␣ medications) and the per-
centage improvement in the active joint count at 6 or 12
months.
Subjects with active arthritis and those with inac-
tive arthritis at the time that the blood sample was
obtained were compared for their initial percentage of
improvement in active joint counts at 6 months and at 12
months. Among the subjects receiving MTX, those with
inactive arthritis at the time of the study visit had a mean
percentage improvement in the active joint count of
80.2% at 6 months after treatment initiation, compared
with a mean percentage improvement of 31.1% in those
with active disease at the time of the study visit (P ⬍
0.0001). When 12-month outcomes in relation to MTX
treatment were compared, again those subjects with
inactive disease at the time of our cross-sectional analy-
sis had a higher mean percentage improvement in the
active joint count at 12 months than did those who had
currently active disease (90% improvement versus 59%
280
Table 3. Concentrations of each MTXGlu subtype according to cluster*
Cluster 1 Cluster 2
(n ⫽ 50) (n ⫽ 39)
MTXGlu1 11.5 ⫾ 8.3 22.6 ⫾ 13.7
MTXGlu2 8.8 ⫾ 4.7 14.4 ⫾ 4.0
MTXGlu3 17.6 ⫾ 9.0 45.0 ⫾ 8.7
MTXGlu4 5.7 ⫾ 5.4 16.7 ⫾ 6.7
MTXGlu5 2.0 ⫾ 2.6 4.9 ⫾ 3.4
* Clusters were determined using the K-means clustering method, with results reported as the mean ⫾ SD
nmoles/liter for each subtype of methotrexate polyglutamate (MTXGlu). Cluster associations were
assessed using the F test (with probabilities reported in parentheses), demonstrating distinct differences
between all 3 clusters or between certain groups and the remaining groups, as indicated.
† Clusters 1 and 3 versus clusters 2 and 3.
‡ Cluster 1 versus clusters 2 and 3.
improvement; P ⬍ 0.0001). When we evaluated these
relationships further, the mean percentage improvement
in the active joint count at 6 months remained signifi-
cantly associated with the presence of inactive disease at
the study visit, even after controlling for MTX dose,
duration of MTX, route of MTX, NSAID use, anti-
TNF␣ use, initial active joint count, and percentage
improvement in active joint count at 12 months (ad-
justed P ⫽ 0.0006, R2 ⫽ 30.2).
Distribution of MTXGlu subtypes. A full descrip-
tion of MTXGlu subtype variability in JIA is discussed
in detail elsewhere (12). To investigate the polyglutama-
tion patterns specific to each MTXGlu species, K-means
clustering of the raw concentrations of each MTXGlu
subtype was performed, and the results revealed 3
distinct clusters. Cluster 1 (n ⫽ 50) had relatively low
concentrations of all subtypes (MTXGlu1–5) (mean ⫾
SD concentration of MTXGlutotal 45.6 ⫾ 19.2 nmoles/
liter), cluster 2 (n ⫽ 39) had moderate concentrations of
intracellular MTXGlu1–5 (mean ⫾ SD concentration of
MTXGlutotal 103.6 ⫾ 20.9 nmoles/liter) but relatively
high concentrations of native MTXGlu1, and cluster 3
(n ⫽ 15) had relatively high concentrations of
MTXGlu1–5 (mean ⫾ SD concentration of MTXGlutotal
169.1 ⫾ 29.6 nmoles/liter) but significantly higher con-
centrations of MTXGlu3–5 (for comparison of each
subtype concentration between clusters, P ⬍ 0.0001)
(Table 3).
Variables associated with MTXGlu clusters.
Clinical variables were assessed for associations with the
above-described K-means clusters. Multivariable logistic
regression analysis controlling for clinical variables that
were observed to be significant (P ⬍ 0.1) in bivariable
analyses revealed that the route of administration, dose,
and duration of MTX remained associated (R2 ⫽ 0.24)
with cluster distribution (P ⬍ 0.0001, P ⫽ 0.034, and P ⫽
0.0007 for cluster 1, cluster 2, and cluster 3, respectively).
BECKER ET AL
Cluster 3
(n ⫽ 15) F (P)
18.5 ⫾ 10.2 11.6 (⬍0.0001)†
15.4 ⫾ 5.2 22.6 (⬍0.0001)‡
80.4 ⫾ 14.6 254.7 (⬍0.0001)
39.5 ⫾ 8.8 160.6 (⬍0.0001)
15.3 ⫾ 8.2 60.2 (⬍0.0001)
For example, the subjects in cluster 3, the group with the
highest concentrations of MTXGlu3–5, were treated with
MTX exclusively by the subcutaneous route and re-
ceived the largest doses of MTX (mean ⫾ SD dosage
0.69 ⫾ 0.25 mg/kg/week) for the shortest period of time
(median 11 months, IQR 6–20 months). Alternatively,
patients in cluster 1, who had relatively low concentra-
tions of all MTXGlu subtypes, had the longest length of
MTX treatment, for a median of 27 months (IQR
10.5–55.3 months), which was significantly different
from that in cluster 2 (P ⫽ 0.006) and cluster 3 (P ⫽
0.003).
Moreover, in comparison with that in cluster 2
and cluster 3, the dose of MTX administered was
significantly lower for the subjects in cluster 1, who
received a mean ⫾ SD dosage of 0.43 ⫾ 0.22 mg/kg/
week (P ⫽ 0.021 versus cluster 2 and P ⫽ 0.0003 versus
cluster 3); there was no statistically significant difference
in dose between cluster 2 and cluster 3 (P ⫽ 0.052).
There was no association between the current use of
folic acid supplements, NSAIDs, corticosteroids, or bio-
logic agents and cluster assignment. Thus, the observed
pattern of MTX polyglutamation appears to be influ-
enced by dose, route of administration, and duration of
MTX treatment.
To determine the contribution of genetic varia-
tion to the observed patterns of MTX polyglutamation,
selected SNPs in genes involved in cellular folate ho-
meostasis and MTX metabolism and utilization were
evaluated for associations with MTXGlu clusters. All
SNPs included in the analyses (34 total) were in Hardy-
Weinberg equilibrium. Initially, a simple bivariable chi-
square analysis was performed to assess associations
between individual SNPs and the K-means clusters. This
analysis identified only a marginal association with 1
SNP, a TYMS variant (6-nucleotide deletion at the
3⬘-untranslated region, rs11280056; P ⫽ 0.042), although
GENOTYPE EFFECTS ON MTXGlu VARIABILITY AND DRUG RESPONSE IN JIA 281

Figure 1. A, Distribution of methotrexate polyglutamate subtypes 1–5 (MTXGlu1–5) clustering in 2 loci,

ADORA2a (rs3761422) and ATIC (rs4673990). Values over columns are the number of alleles for ATIC, and
values beside rows are the number of alleles for ADORA2a. In each cell, open bars (and values over the bars)
represent the number of subjects in cluster 1 (low concentration of red blood cell [RBC] MTXGlu1–5), while solid
bars represent the number of subjects in cluster 3 (high concentration of RBC MTXGlu1–5). Dark gray–shaded
cells indicate the genotype combinations in which the likelihood of being in cluster 1 is relatively low, while light
gray–shaded cells indicate the genotype combinations in which the likelihood of being in cluster 3 is relatively low.
For this 2-loci model, the genetic prediction has a balanced accuracy of 80.3%, sensitivity of 74.0%, specificity of
86.7%, and precision of 99.5%. B, Radical diagram of gene–gene interactions in 3-locus multifactor dimension-
ality reduction models. The contribution of each gene (gene effects) was measured by entropy-based information
gain. Values for individual gene effects are shown in the rectangles for each single-nucleotide polymorphism
(SNP), while values for joint effects are shown in the lines connecting 2 corresponding SNPs.

this association was not significant after correction for ysis, the TYMS 6-nucleotide deletion (rs11280056) re-
multiple comparisons. mained associated with cluster outcome, whereas ATIC
To test the hypothesis that the variability in (rs4673990) and ADORA2a (rs3761422) did not show
patterns of MTX polyglutamation was a function of
epistasis, or gene–gene interactions, between folate
pathway genes, the MDR method was applied to the
Table 4. Multifactor dimensionality reduction analysis evaluating
clusters with the lowest and highest MTXGlu concen- gene–gene interactions that could differentiate cluster 1 from cluster
trations (extreme clusters 1 and 3, respectively). The 3*
most pronounced combination of SNPs differentiating Sensitivity measure Value
subjects in cluster 1 from those in cluster 3 was ATIC
(SNP rs4673990) and ADORA2a (SNP rs3761422) (Fig- Accuracy, %
Training data set 80.3
ures 1A and B). This two-way model presented the best Testing data set 80.3
overall performance, with a testing accuracy of 80.3% (P CVC 10/10
⫽ 0.006) and a cross-validation consistency of 10/10 (P ⫽ OR (95% CI) 18.5 (3.67–93.23)
P
0.0550) (Table 4). Subjects who carried this SNP com- Testing data accuracy 0.0060
bination were 18.5 times more likely to be in cluster 1 as CVC 0.0550
compared with cluster 3 (odds ratio 18.5, 95% confi- OR ⬍0.0001
dence interval 3.7–93.2; P ⬍ 0.0001). Sensitivity was * Cluster 1 was defined as subjects having low methotrexate polyglu-
74%, specificity was 86.7%, and overall accuracy was tamate (MTXGlu) concentrations of red blood cell (RBC)
MTXGlu1–5, while cluster 3 was defined as those having high concen-
80.3%. trations of RBC MTXGlu1–5, specifically, high MTXGlu3–5. The most
A multivariate logistic regression model was built pronounced combination of single-nucleotide polymorphisms differ-
with inclusion of the 3 SNPs (genotypes) that were entiating subjects in cluster 1 from those in cluster 3 was in the model
evaluating ATIC (SNP rs4673990, A⬎G) and ADORA2a (SNP
observed to be associated with cluster outcome in the rs3761422). CVC ⫽ cross-validation count; OR ⫽ odds ratio; 95%
chi-square and MDR analyses. In this multivariate anal- CI ⫽ 95% confidence interval.
282
any statistically significant associations with cluster out-
come (P ⫽ 0.27 and P ⫽ 0.70, respectively). When
clinical variables that were significantly associated with
cluster outcome were added to the model, we found that
route of administration (P ⬍ 0.0001), dose (P ⫽ 0.043),
and duration of MTX treatment (P ⫽ 0.001), in addition
to the presence of the TYMS SNP (P ⫽ 0.021), were
associated with cluster (R2 ⫽ 0.31).
MTXGlu polyglutamation patterns (clusters)
and clinical outcomes. There was no association be-
tween cluster assignment and the presence of joints with
active arthritis at the study visit, the percentage change
in active joint count at 6 or 12 months, or the occurrence
of GI side effects. There were, however, differences
between groups in the risk of an elevated liver enzyme
value on LFTs. Cluster 1 (having the lowest MTXGlu
concentrations) had the lowest risk of having elevated
levels on LFTs at the time of the visit. Abnormalities on
LFTs were observed in 4% of subjects in cluster 1 as
compared with 23% in cluster 2 and 27% in cluster 3
(each P ⫽ 0.014 versus cluster 1).
Further analyses of long-chain MTXGlu3–5 were
completed. Long-chain polyglutamate concentrations,
corrected for the dose of MTX administered, were
higher in subjects who had elevated liver enzyme levels
on LFTs at the time of their visit (mean ⫾ SD concen-
tration of MTXGlu3–5 173.0 ⫾ 162.9 nmoles/liter versus
111.8 ⫾ 85.5 nmoles/liter in those with no abnormal
findings; P ⫽ 0.03) and in subjects who reported GI side
effects (mean ⫾ SD concentration of MTXGlu3–5
159.2 ⫾ 134.4 nmoles/liter versus 107.7 ⫾ 85.2 nmoles/
liter in those without GI side effects; P ⫽ 0.013). No
associations between route of administration and ele-
vated LFT values (P ⫽ 0.14) or GI side effects (P ⫽
0.15) were noted. In addition, there were no associations
between dose of MTX received and elevated levels on
LFTs (P ⫽ 0.35) or GI side effects (P ⫽ 0.11).
In assessing the outcome of active arthritis, there
was no association between individual MTXGlu sub-
types, or specifically of long-chain polyglutamates, with
favorable outcome, namely, lack of active arthritis. Ad-
ditionally, there was no association between route and
dose of MTX with the presence of active arthritis at the
time that the study sample was obtained.
DISCUSSION
Although MTX is widely administered by adult
and pediatric rheumatologists to treat inflammatory
arthritis, there is extensive variability in response to the
drug (2,3,18–23). Attempts to better understand this
BECKER ET AL
variability or, better yet, to predict the clinical response
have led to the measurement of intracellular MTXGlu
concentrations as a biomarker of drug effect. Intracel-
lular MTXGlu concentrations are more stable than
serum levels of MTX (24), and therefore are more likely
to be associated with treatment outcome. Long-chain
MTXGlu concentrations have been associated with im-
proved response in adults with RA (7,8,25), but recently,
there have been inconsistencies in the published data
(26). In children, we have reported a wide variability in
intracellular MTXGlu concentrations in those with JIA,
with a strong association between route of administra-
tion and MTXGlu chain length (12). To explore factors
contributing to this variability and the consequences for
clinical response, we herein evaluated the clinical and
genetic predictors of MTXGlu patterns, as well as the
associations between MTXGlu and outcomes in patients
with JIA.
The K-means cluster methods applied in this
analysis revealed 3 distinct clusters of MTXGlu, and
dose, duration of MTX use, and route of administration
all contributed to the polyglutamation pattern. One
explanation for this observation may be that the clusters
are populated with patients with varying levels of disease
severity. For example, subjects in cluster 3 were receiv-
ing the highest doses of MTX, for the shortest period of
time administered via the subcutaneous route, which
may be explained if this cluster is populated with the
most severely affected patients. There were, however, no
differences in the number of joints with active arthritis at
presentation (P ⫽ 0.99), at 6 months (P ⫽ 0.22), or at 12
months (P ⫽ 0.43) between groups.
Additionally, there were no significant differ-
ences in JIA subtype, age, or the use of folic acid
supplements between groups. However, less improve-
ment in the active joint count at 6 months in cluster 3
(mean ⫾ SD 17 ⫾ 74% improvement) was noted
compared with that in cluster 2 (56 ⫾ 50.4% improve-
ment) and cluster 1 (59.5 ⫾ 60.8% improvement) (P ⫽
0.049), and subjects in cluster 3 had a higher likelihood
of having initially taken prednisone and having been
prescribed a biologic agent by 6 months, in comparison
with the other groups, suggesting a slower response to
therapy. However, the percentage improvement in the
active joint count at 12 months was not significantly
different between groups (P ⫽ 0.97), suggesting that,
although cluster 3 may not have had as rapid an im-
provement as the other groups, by 12 months that
difference appears to have been less significant.
Our observations showing that the cluster com-
prising subjects treated with MTX for the longest period
GENOTYPE EFFECTS ON MTXGlu VARIABILITY AND DRUG RESPONSE IN JIA
of time had the lowest concentrations of MTXGlu, while
the cluster comprising subjects treated with MTX for the
shortest period of time had the highest concentrations of
MTXGlu are the opposite of what one might expect
when considering the effect of length of treatment on
achieving steady state for the longer-chain MTXGlu
(11). We may have captured a previously underrecog-
nized influence of noncompliance in the cohort of
patients with longstanding JIA; however, these patients
were also receiving lower weekly MTX doses than were
the patients in the other 2 clusters (mean ⫾ SD 0.43 ⫾
0.22 mg/kg/week compared with 0.55 ⫾ 0.24 mg/kg/week
and 0.69 ⫾ 0.25 mg/kg/week in cluster 2 and cluster 3,
respectively). Small numbers in cluster 3 (n ⫽ 15) may
also have resulted in a lack of statistical power, which
may explain the lack of difference seen between groups
in some of the above-mentioned clinical variables. Prac-
titioner style or preference for MTX dose and route of
administration, as well as a more recent focus on early
and aggressive treatment, may also potentially explain
the differences seen between clusters. These hypotheses
can only be tested more clearly with a larger prospective,
controlled study.
In an attempt to better elucidate the genetic
contribution to variation and patterns in intracellular
MTXGlu, we utilized MDR to investigate the role of
gene–gene interactions (epistasis). SNPs within the pu-
rine (ATIC) and adenosine (ADORA2a) pathways were
found to differentiate clusters with high concentrations
(specifically, MTXGlu3–5) from those with low overall
concentrations. Additionally, with the use of more tra-
ditional statistical methods, we identified a SNP in
TYMS that was associated with cluster outcome as well.
We do not believe that these genes have a direct effect
on MTXGlu concentrations or patterns of polyglutama-
tion. Rather, we hypothesize that the observed patterns
of polyglutamation reflect the underlying cellular folate
status. The folate pathway is responsible for purine
synthesis, pyrimidine synthesis, adenosine metabolism,
and homocysteine remethylation. In the presence of
MTX and the resulting folate depletion that is the
intended therapeutic goal, the cell must allocate dwin-
dling folate resources to specific, as yet unknown sites,
and therefore a better understanding of the variation
(SNPs) in checkpoint genes may help to determine how
a pharmacologically folate-depleted cell allocates its
resources and slows down some processes to accommo-
date for other needs.
Allelic variation in the genes detected, TYMS,
ATIC, and ADORA2a, may result in more profound
inhibition of enzyme activity with reduced formation of
283
products, which, through feedback mechanisms, may be
perceived as a more profound folate-depleted state.
However, we cannot fully answer these questions with-
out a better understanding of the intracellular concen-
trations of individual folate redox states and folate
polyglutamates. Therefore, in light of the knowledge
that these genes could play important roles in the
multiple processes linked to the folate pathway, we
believe that the relevance of the genes will further
increase with our increasing knowledge about the mech-
anisms of action of MTX. In support of this notion,
recent findings have also suggested that SNPs in the
purine synthesis pathway may be linked to MTX re-
sponse in children with JIA (27) as well as those with
leukemia (28). Moreover, the results in the present study
may also suggest that these SNPs could play a role in
MTX biotransformation and response in JIA.
Variability in MTXGlu concentrations, however,
has little value if there are no associations between
MTXGlu concentrations and clinical outcome. In con-
trast to the observations reported by Dervieux and
colleagues (7–9), and in agreement with those of
Dolezalova and colleagues (29), we found no association
between intracellular MTXGlu concentrations and the
presence of active arthritis. We did, however, see a trend
toward higher concentrations of short-chain polygluta-
mates, MTXGlu1 ⫹ 2, in subjects with inactive arthritis,
which, interestingly, is in contrast to the findings re-
ported in the literature in studies of adult patients (7–9).
However, within this cohort of patients with JIA,
we did find associations between MTXGlu and drug
toxicity. The group with the lowest intracellular MTX-
Glu concentrations had the lowest risk of liver enzyme
elevation, and the group with the highest concentration
of long-chain MTXGlu3–5 had the highest risk of liver
enzyme elevation. We have previously noted a higher
risk of MTX toxicity in the liver in subjects with JIA who
received higher doses of MTX (3); therefore, it initially
may not be surprising that this association was found,
since cluster 3 was also the group who were receiving the
highest doses of MTX. However, even after correcting
MTXGlu concentrations for the dose of MTX, higher
concentrations of long-chain polyglutamates were seen
in subjects with liver enzyme elevation and reported GI
side effects. These children may have inherent differ-
ences in how they intracellularly transform MTX, which
may put them at risk for adverse effects from the drug.
SNP combinations that may be used to predict
cluster subtype may also highlight patients who will
respond to MTX differently; for example, if SNPs within
ATIC and the adenosine receptor ADORA2a will de-
284
crease the ability of the cell to adequately produce or
respond to adenosine, these patients may have an atten-
uated antiinflammatory response to the drug, and thus
they would have a higher likelihood of requiring higher
doses of MTX to control their clinical disease. Not one
single factor can explain the MTXGlu variability and
associations with outcomes, and we therefore need to
determine the intracellular folate concentrations and
polyglutamate concentrations to complete the picture.
Nevertheless, this is the first study to describe an asso-
ciation between MTXGlu concentrations and adverse
outcomes in relation to MTX treatment in patients with
JIA.
We acknowledge that our study had some limita-
tions. It was designed as a cross-sectional analysis of
patients with JIA at a single center, and thus did not
allow us to dynamically follow up trends or effectively
measure the response to MTX over time. The purpose
of conducting a cross-sectional and retrospective analy-
sis initially was to help us better frame the hypotheses
and design the prospective studies to test these hypoth-
eses, which are currently in development. Our measures
of clinical response, including the presence and number
of joints with active disease and the ESR, are not as
rigorous as outcome measurements such as the ACR
Pedi 30 or the Disease Activity Score. Since inactive
disease or remission related to medications is the de-
sired outcome of therapy in JIA, and since we were
limited by the inherent design of this pilot study, we
chose to evaluate the presence or absence of disease
activity, which, some may argue, is more clinically rele-
vant.
The improvement in active joint counts at 6 and
12 months after therapy initiation helped to strengthen
our speculations that patients who were categorized as
responders and showed a lack of active arthritis at the
time that MTXGlu was measured had a clinical course
that was more responsive to MTX. The data presented
herein suggest that those without active arthritis at the
time that the study sample was obtained had a superior
early response to MTX, and therefore improvement at 6
months may be an important factor in determining
overall response. Although, through a cross-sectional
design, we were unable to control for variables such as
MTX dose and route of administration, this relatively
large cohort of patients with JIA has provided us with
the preliminary data and methods to develop more
rigorous, well-controlled prospective studies. These data
show a relationship between MTXGlu and outcomes in
JIA, which may fuel further studies in this area.
The dosing strategy for MTX in JIA remains
BECKER ET AL
limited, since it is necessary to await the eventual
response before effectiveness or toxicity can be deter-
mined. We continue to search for better and more
reliable markers that could be used to predict who will
have a favorable response or poor response to the drug
early in the course of therapy. With the prospective
studies that are being developed, we hope to gain a
better understanding of an individual’s baseline risk and
ability to adapt to folate disturbances with MTX ther-
apy.
ACKNOWLEDGMENTS
The authors would like to gratefully acknowledge
Jaylene Weigel and Chelsey Palmer for their assistance with
patient recruitment and sample preparation, as well as Todd
D. Williams and the University of Kansas Mass Spectrometry
facility for their guidance and collaboration in this project.
AUTHOR CONTRIBUTIONS
All authors were involved in drafting the article or revising it
critically for important intellectual content, and all authors approved
the final version to be published. Dr. Becker had full access to all of the
data in the study and takes responsibility for the integrity of the data
and the accuracy of the data analysis.
Study conception and design. Becker, Stobaugh, Leeder.
Acquisition of data. Becker, Lasky, Hoeltzel.
Analysis and interpretation of data. Becker, Gaedigk, van Haandel,
Thomas, Dai, Stobaugh, Leeder.
REFERENCES
1. Cassidy JT, Petty RE, Laxer RM, Lindsley CB, editors. Textbook
of pediatric rheumatology. 5th ed. Philadelphia: Elsevier Saun-
ders; 2005.
2. Ortiz-Alvarez O, Morishita K, Avery G, Green J, Petty RE,
Tucker LB, et al. Guidelines for blood test monitoring of metho-
trexate toxicity in juvenile idiopathic arthritis. J Rheumatol 2004;
31:2501–6.
3. Becker ML, Rose CD, Cron RQ, Sherry DD, Bilker WB, Laut-
enbach E. Effectiveness and toxicity of methotrexate in juvenile
idiopathic arthritis: comparison of two initial dosing regimens.
J Rheumatol 2010;37:870–5.
4. Chabner BA, Allegra CJ, Curt GA, Clendeninn NJ, Baram J,
Koizumi S, et al. Polyglutamation of methotrexate: is methotrexate
a prodrug? J Clin Invest 1985;76:907–12.
5. Cronstein BN, Naime D, Ostad E. The antiinflammatory mecha-
nism of methotrexate: increased adenosine release at inflamed
sites diminishes leukocyte accumulation in an in vivo model of
inflammation. J Clin Invest 1993;92:2675–82.
6. Allegra CJ, Chabner BA, Drake JC, Lutz R, Rodbard D, Jolivet J.
Enhanced inhibition of thymidylate synthase by methotrexate
polyglutamates. J Biol Chem 1985;260:9720–6.
7. Dervieux T, Furst D, Lein DO, Capps R, Smith K, Walsh M, et al.
Polyglutamation of methotrexate with common polymorphisms in
reduced folate carrier, aminoimidazole carboxamide ribonucleo-
tide transformylase, and thymidylate synthase are associated with
methotrexate effects in rheumatoid arthritis. Arthritis Rheum
2004;50:2766–74.
8. Dervieux T, Furst D, Lein DO, Capps R, Smith K, Caldwell J, et
GENOTYPE EFFECTS ON MTXGlu VARIABILITY AND DRUG RESPONSE IN JIA
al. Pharmacogenetic and metabolite measurements are associated
with clinical status in patients with rheumatoid arthritis treated
with methotrexate: results of a multicentred cross sectional obser-
vational study. Ann Rheum Dis 2005;64:1180–5.
9. Dervieux T, Greenstein N, Kremer J. Pharmacogenomic and
metabolic biomarkers in the folate pathway and their association
with methotrexate effects during dosage escalation in rheumatoid
arthritis. Arthritis Rheum 2006;54:3095–103.
10. Schroder H, Fogh K. Methotrexate and its polyglutamate deriva-
tives in erythrocytes during and after weekly low-dose oral meth-
otrexate therapy of children with acute lymphoblastic leukemia.
Cancer Chemother Pharmacol 1988;21:145–9.
11. Dalrymple JM, Stamp LK, O’Donnell JL, Chapman PT, Zhang M,
Barclay ML. Pharmacokinetics of oral methotrexate in patients
with rheumatoid arthritis. Arthritis Rheum 2008;58:3299–308.
12. Becker ML, van Haandel L, Gaedigk R, Lasky A, Hoeltzel M,
Stobaugh J, et al. Analysis of intracellular methotrexate polyglu-
tamates in patients with juvenile idiopathic arthritis: effect of route
of administration on variability in intracellular methotrexate poly-
glutamate concentrations. Arthritis Rheum 2010;62:1803–12.
13. Petty RE, Southwood TR, Manners P, Baum J, Glass DN,
Goldenberg J, et al. International League of Associations for
Rheumatology classification of juvenile idiopathic arthritis: second
revision, Edmonton, 2001. J Rheumatol 2004;31:390–2.
14. Van Haandel L, Becker ML, Leeder JS, Williams TD, Stobaugh
JF. A novel high-performance liquid chromatography/mass spec-
trometry method for improved selective and sensitive measure-
ment of methotrexate polyglutamation status in human red blood
cells. Rapid Commun Mass Spectrom 2009;23:3693–702.
15. Giannini EH, Ruperto N, Ravelli A, Lovell DJ, Felson DT,
Martini A. Preliminary definition of improvement in juvenile
arthritis. Arthritis Rheum 1997;40:1202–9.
16. Hahn LW, Ritchie MD, Moore JH. Multifactor dimensionality
reduction software for detecting gene–gene and gene–
environment interactions. Bioinformatics 2003;19:376–82.
17. Giannini EH, Ruperto N, Ravelli A, Lovell DJ, Felson DT,
Martini A. Preliminary definition of improvement in juvenile
arthritis. Arthritis Rheum 1997;40:1202–9.
18. Ruperto N, Murray KJ, Gerloni V, Wulffraat N, de Oliveira SK,
Falcini F, et al, for the Pediatric Rheumatology International
Trials Organization. A randomized trial of parenteral methotrex-
ate comparing an intermediate dose with a higher dose in children
with juvenile idiopathic arthritis who failed to respond to standard
doses of methotrexate. Arthritis Rheum 2004;50:2191–201.
19. Lambert CM, Sandhu S, Lochhead A, Hurst NP, McRorie E,
Dhillon V. Dose escalation of parenteral methotrexate in active
285
rheumatoid arthritis that has been unresponsive to conventional
doses of methotrexate: a randomized, controlled trial. Arthritis
Rheum 2004;50:364–71.
20. Cohen S, Cannon GW, Schiff M, Weaver A, Fox R, Olsen N, et al.
Two-year, blinded, randomized, controlled trial of treatment of ac-
tive rheumatoid arthritis with leflunomide compared with metho-
trexate. Arthritis Rheum 2001;44:1984–92.
21. Wallace CA, Sherry DD. Preliminary report of higher dose
methotrexate treatment in juvenile rheumatoid arthritis. J Rheu-
matol 1992;19:1604–7.
22. Alsufyani K, Ortiz-Alvarez O, Cabral DA, Tucker LB, Petty RE,
Malleson PN. The role of subcutaneous administration of metho-
trexate in children with juvenile idiopathic arthritis who have failed
oral methotrexate. J Rheumatol 2004;31:179–82.
23. Reiff A, Shaham B, Wood BP, Bernstein BH, Stanley P, Szer IS.
High dose methotrexate in the treatment of refractory juvenile
rheumatoid arthritis. Clin Exp Rheumatol 1995;13:113–8.
24. Stamp LK, O’Donnell JL, Chapman PT, Zhang M, Frampton C,
James J, et al. Determinants of red blood cell methotrexate
polyglutamate concentrations in rheumatoid arthritis patients re-
ceiving long-term methotrexate treatment. Arthritis Rheum 2009;
60:2248–56.
25. Dervieux T, Kremer J, Lein DO, Capps R, Barham R, Meyer G, et
al. Contribution of common polymorphisms in reduced folate
carrier and gamma-glutamylhydrolase to methotrexate polygluta-
mate levels in patients with rheumatoid arthritis. Pharmacogenet-
ics 2004;14:733–9.
26. Stamp LK, O’Donnell JL, Chapman PT, Zhang M, James J,
Frampton C, et al. Methotrexate polyglutamate concentrations are
not associated with disease control in rheumatoid arthritis patients
receiving long-term methotrexate therapy. Arthritis Rheum 2010;
62:359–68.
27. Hinks A, Moncrieffe H, Martin P, Lal SD, Ursu S, Kassoumeri L,
et al. Genetic polymorphisms in key methotrexate (MTX) pathway
genes associated with response to MTX treatment in juvenile
idiopathic arthritis [abstract]. Arthritis Rheum 2009;60 Suppl:
S233.
28. Stocco G, Cheok MH, Crews KR, Dervieux T, French D, Pei D,
et al. Genetic polymorphism of inosine triphosphate pyrophos-
phatase is a determinant of mercaptopurine metabolism and
toxicity during treatment for acute lymphoblastic leukemia. Clin
Pharmacol Ther 2009;85:164–72.
29. Dolezalova P, Krijt J, Chladek J, Nemcova D, Hoza J. Adenosine
and methotrexate polyglutamate concentrations in patients with
juvenile arthritis. Rheumatology (Oxford) 2005;44:74–9.